Combining generative aligners and transition-based parsers

A system using posterior distributions and stochastic oracle policies mitigates error propagation in transition-based parsers, improving robustness and generalization by exposing the parser to multiple models and preventing over-fitting.

US12639554B2Active Publication Date: 2026-05-26MASSACHUSETTS INST OF TECH +1
View PDF 32 Cites 0 Cited by

Patent Information

Authority / Receiving Office
US · United States
Patent Type
Patents(United States)
Current Assignee / Owner
MASSACHUSETTS INST OF TECH
Filing Date
2023-04-27
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Error propagation from inaccuracies in generative alignment to transition-based parser training leads to over-fitting and reduced robustness, particularly in spoken language understanding tasks.

Method used

Implement a system that computes a posterior distribution over hard alignments using a generative alignment component, applies a stochastic oracle policy, and incorporates a scaling factor to mitigate error propagation, thereby exposing the parser to multiple explanatory models and preventing over-fitting.

Benefits of technology

The system effectively reduces error propagation and over-fitting, enhancing the parser's robustness and generalization capabilities by propagating uncertainty from the generative stage to the parser training stage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure US12639554-D00000_ABST
    Figure US12639554-D00000_ABST
Patent Text Reader

Abstract

Systems, computer-implemented methods, and computer program products to facilitate reducing error propagation when combining generative aligners and transition-based aligners are provided. According to an embodiment, a system can comprise a processor that executed components stored in memory. The computer executable components comprise a generative alignment component, an error propagation component, a discriminative parser, and a stochastic oracle policy component. The error propagation component can compute a posterior distribution over one or more hard alignments of parts given a pair of the generative alignment component. The discriminative parser can be trained via the stochastic oracle policy component to reduce error propagation when combining the generative alignment component with the discriminative parser.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Generating dependency parses of text segments using neural networks

    US10878188B2

  • Determining relatedness of data using graphs to support machine learning, natural language parsing, search engine, or other functions

    US11238095B1

  • Search engine with natural language-based robust parsing of user query and relevance feedback learning

    US20040243568A1

  • Training a statistical parser on noisy data by filtering

    US20060277028A1

  • Generating word embeddings with a word embedder and a character embedder neural network models

    US20220083837A1